01
The classroom
Fourteen years correcting the same L1-interference patterns by hand — the /z/→/dʒ/ swaps, the /v/~/w/ merges. Vaaani’s cause-net encodes exactly those patterns, per home language. This came from classrooms, not a paper.
← Back
Who builds it
Built by someone who has corrected these errors by hand.
Neil Shankar Ray — applied linguist and AI engineer. 14 years teaching English across Indian schools, to L1-Bangla, Hindi and Marathi students. MA in Applied Linguistics; AI/ML certification at IIT Patna; an early Diploma in Software Engineering from NIIT.
Three lenses on one problem: making a machine reason about language the way a linguist does — and prove it.
02
The linguistics
An MA in Applied Linguistics. Vaaani treats language as a linguist would — a graph of sounds, morphemes and meanings with real relationships — not a bag of tokens, and not a prompt.
03
The engineering
A Diploma in Software Engineering and an AI/ML certification at IIT Patna. The teaching decisions run on explicit probabilistic reasoning over the language graph — deterministic, auditable, on CPU — not a large language model.
I built Vaaani because the tutors I saw optimised for a fast answer, not a child who understands. And they were black boxes — you could never say why a decision was made. So I built the opposite: an engine that models each child, and can show its reasoning to a parent, a teacher, or a board.
See it reason on your own students.
A single cohort, real children, measured results you can read line by line. We’ll set up your pilot personally.